The growing demand for personalized fashion experiences has highlighted the limitations of existing virtual wardrobe and recommendation systems, which often address only one aspect—either garment visualization or style prediction— resulting in incomplete solutions. However, many existing platforms face challenges in preserving texture fidelity, ensuring alignment across diverse body poses, and scaling effectively to real-world scenarios. To address these challenges, this work proposes the Smart Virtual Wardrobe, an AI-powered platform that integrates garment classification, outfit planning, and photorealistic virtual try-on within a unified framework. The system employs a fine-tuned ResNet-34 model, trained on a large-scale fashion dataset, to automatically classify clothing items with 94.9% accuracy. In parallel, a virtual try-on module based on EfficientNet and trained on the Kaggle VITON dataset achieves 98% accuracy, enabling realistic visualization of apparel combinations on user images. The platform further enhances personalization by providing outfit recommendations based on season, event type, weather, and style preferences. The system is deployed with a React front-end and a Python FastAPI backend, enabling a scalable, interactive, and context-aware solution for next-generation fashion management and retail applications.
Paper
Full text
Smart Virtual Wardrobe: AI-Powered Outfit Planner and Style Assistant
Semantic Scholar · 2025
Abstract
The growing demand for personalized fashion experiences has highlighted the limitations of existing virtual wardrobe and recommendation systems, which often address only one aspect—either garment visualization or style prediction— resulting in incomplete solutions. However, many existing platforms face challenges in preserving texture fidelity, ensuring alignment across diverse body poses, and scaling effectively to real-world scenarios. To address these challenges, this work proposes the Smart Virtual Wardrobe, an AI-powered platform that integrates garment classification, outfit planning, and photorealistic virtual try-on within a unified framework. The system employs a fine-tuned ResNet-34 model, trained on a large-scale fashion dataset, to automatically classify clothing items with 94.9% accuracy. In parallel, a virtual try-on module based on EfficientNet and trained on the Kaggle VITON dataset achieves 98% accuracy, enabling realistic visualization of apparel combinations on user images. The platform further enhances personalization by providing outfit recommendations based on season, event type, weather, and style preferences. The system is deployed with a React front-end and a Python FastAPI backend, enabling a scalable, interactive, and context-aware solution for next-generation fashion management and retail applications.